A parallel control method and system for partially unknown system models

Abstract

This invention proposes a parallel control method and system for systems with partially unknown system models. The method includes constructing an RBF neural network to approximate the unknown parts of the system model, reconstructing the system model to obtain a reconstructed system model; constructing a sliding mode function based on the reconstructed system model; and constructing a parallel controller based on the RBF neural network and the sliding mode function to control the system to an equilibrium state. This invention utilizes an RBF neural network to approximate the unknown parts of the system model to obtain a complete reconstructed system model and designs a corresponding parallel controller to control the system. This effectively solves the problem of difficulty in generating control signals when the system state cannot be obtained due to partially unknown system models, and can be widely applied to practical systems with partially unknown system models.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control, and more specifically, to a parallel control method and system for partially unknown system models. Background Art

[0002] In today's industrial control systems, most system control problems are analyzed through state feedback control methods, that is, by designing a state feedback controller to form a closed-loop system, the control law is used as a function of the system state to analyze the system control problem. Traditional state feedback controllers are only related to the system state, but not to the nature of the controller, resulting in a large change in the control signal with the system state, which brings great difficulties to the execution of the controller. In addition, the control signal is generated passively, and it is difficult to generate a control signal when the system state cannot be obtained.

[0003] There is a parallel control method for intelligent workshops. It builds a parallel control simulation platform, establishes a parallel execution mechanism, and corrects and optimizes the parallel control system to perform parallel control on the intelligent workshop system. It generates a stable control signal by combining the workshop system status and control status, and realizes real-time monitoring and three-dimensional visualization of the workshop operation status and real-time control of the workshop performance. However, the above method is aimed at a complete and accurate linear system model, and cannot be directly applied to a system model with unknown parts of the system, which has great limitations in practical applications. Summary of the invention

[0004] The present invention aims to solve the defect that the existing parallel control method cannot be directly applied to a system in which the system model is partially unknown and has great application limitations.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, the present invention proposes a parallel control method for a partially unknown system model, comprising the following steps:

[0007] S1: constructing an RBF neural network, using the RBF neural network to approximate the unknown part of the system model, reconstructing the system model, and obtaining a reconstructed system model;

[0008] S2: constructing a sliding mode function according to the reconstructed system model;

[0009] S3: construct a parallel controller based on the RBF neural network and the sliding mode function, and use the parallel controller to control the system to run to a balanced state.

[0010] As a preferred solution, in S1, an RBF neural network is constructed for the unknown part of the system model, specifically including: determining the number of nodes in the input layer, hidden layer and output layer of the RBF neural network for the unknown part of the system model, selecting a Gaussian basis function as an activation function, and setting the weight matrix between the input layer and the hidden layer to a unit matrix to obtain the RBF neural network.

[0011] As a preferred solution, the unknown part of the system model includes a continuous-time state function.

[0012] As a preferred solution, in S1, the unknown continuous-time state function of the system model is approximated by the RBF neural network, and the system model is reconstructed to obtain a reconstructed system model Its expression is as follows:

[0013]

[0014] in, is the output of the RBF neural network, representing the unknown continuous-time state function approximated by the RBF neural network; Represents the input of the RBF neural network, which is the n-dimensional system state vector; represents the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network, T represents the transpose of the matrix, h(x) represents the output of the Gaussian basis function of the RBF neural network, is the n-dimensional system control vector.

[0015] As a preferred solution, in S2, according to the reconstructed system model, a sliding mode function s is constructed, and its expression is as follows:

[0016]

[0017] Among them, c is an adjustable parameter that satisfies the Hurwitz condition.

[0018] As a preferred solution, in S3, the parallel controller includes a switching component and a continuous component; wherein the switching component control system moves quickly toward the sliding surface, and the continuous component control system moves along the sliding surface until the system runs to an equilibrium state, and its expression is as follows:

[0019]

[0020]

[0021]

[0022] Where k>0, and

[0023]

[0024] in, is the parallel controller, u is the control signal, represents the continuous component of the parallel controller, The switching component of the parallel controller, for Derivative of , ε bound represents the upper bound of the RBF neural network approximation error, sgn(·) represents the sign function, and k is an adjustable parameter greater than zero.

[0025] As a preferred solution, after obtaining the sliding mode function of the reconstructed system model, it also includes designing the update law of the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network. Using the update law Update the estimate of the ideal weight matrix The update law The expression is as follows:

[0026]

[0027] Among them, γ is a configurable learning rate.

[0028] In a second aspect, the present invention further proposes a parallel control system for a partially unknown system model, which is applied to the parallel control method for a partially unknown system model described in any of the above schemes, comprising:

[0029] RBF neural network building module, used to build RBF neural network;

[0030] A system reconstruction module, used to use the RBF neural network to approximate the unknown part of the system model, reconstruct the system model, and obtain a reconstructed system model;

[0031] A sliding mode function design module, used for constructing a sliding mode function of the reconstruction system model;

[0032] Parallel controller building module, used to build parallel controllers based on RBF neural network and sliding mode function;

[0033] A control module is used to control the system to run to a balanced state using the parallel controller.

[0034] As a preferred solution, the unknown part of the system model is a continuous-time state function.

[0035] In a third aspect, the present invention further proposes a computer system comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program in the memory, it implements the steps of any scheme in the above-mentioned parallel control method for partially unknown system models.

[0036] Compared with the prior art, the beneficial effect of the technical solution of the present invention is: the present invention uses the RBF neural network to approximate the unknown part of the system model, obtains a complete reconstructed system model, and designs a corresponding parallel controller to control the system, so that the system runs to a balanced state, and the closed-loop system tends to be stable, which effectively solves the problem of difficulty in generating control signals when the system state cannot be obtained when the system model is partially unknown, and can be widely used in actual systems where the system model is partially unknown. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the parallel control method for partially unknown system models.

[0038] Figure 2 Schematic diagram of the parallel control method for partially unknown system models.

[0039] Figure 3 The schematic diagram of the parallel controller.

[0040] Figure 4 Architectural diagram of a parallel control system with partially unknown system models. DETAILED DESCRIPTION

[0041] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0042] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0043] Example 1

[0044] This embodiment proposes a parallel control method for partially unknown system models. Figure 1 , Figure 1 The flowchart of the parallel control method for the partially unknown system model includes the following steps:

[0045] S1: constructing an RBF neural network, using the RBF neural network to approximate the unknown part of the system model, reconstructing the system model, and obtaining a reconstructed system model.

[0046] S2: Constructing a sliding mode function of the reconstruction system model.

[0047] S3: A parallel controller is constructed based on the RBF neural network and the sliding mode function, and the parallel controller is used to control the system to run to an equilibrium state. The equilibrium state refers to the state in which the derivative of the system state x is equal to 0, that is, the system does not change after running to this state and reaches stability.

[0048] Since the system model is partially unknown, in order to obtain better performance, it is necessary to use RBF neural network to approximate the unknown part of the system model and reconstruct the system model during the design process of the parallel controller. RBF (Radial Basis Function) neural network has a universal approximation characteristic, which can approximate any nonlinear function in a compact set and with arbitrary precision. The present invention combines the input and output data of the existing system with the RBF neural network, and continuously corrects the weights of the RBF neural network to fit the functional relationship between the input variables and the output variables, so that the error between the predicted output and the actual output is minimized, and the adaptive approximation of the unknown part of the system model is realized.

[0049] In addition, the traditional state feedback controller u=-Kx directly obtains the control signal from the system state, but when the system state is difficult to obtain or cannot be obtained, the control signal cannot be obtained. Therefore, the present invention proposes a parallel control method and designs a corresponding parallel controller. The change of the system control signal is not only related to the system state, but also to the parallel controller itself.

[0050] The present invention utilizes RBF neural network to approximate the unknown part of the system model to obtain a complete reconstructed system model, and designs a corresponding parallel controller to control the system, which effectively solves the problem of difficulty in generating control signals when the system state cannot be obtained when part of the system model is unknown. The present invention can be widely used in system models with unknown parts such as motor control systems, robot control, servo systems and aircraft motion control systems.

[0051] Example 2

[0052] See also Figure 2 , Figure 2 The schematic diagram of the parallel control method for a partially unknown system model is shown in FIG. 1 . The present embodiment provides a parallel control method for a partially unknown system model, comprising the following steps:

[0053] S1: constructing an RBF neural network, using the RBF neural network to approximate the unknown part of the system model, reconstructing the system model, and obtaining a reconstructed system model.

[0054] In this embodiment, the expression of the system model is as follows:

[0055]

[0056] in, is the n-dimensional system state vector, is the n-dimensional system control vector, is the unknown part of the system. In this embodiment, f(x) is an unknown continuous-time state function, and f(0)=0, and f(x) is Lipschitz continuous.

[0057] In this embodiment, for the unknown part of the system model, the number of nodes in the input layer, hidden layer and output layer is determined, and the Gaussian basis function is selected as the activation function. In order to speed up the training of the neural network, the weight matrix of the input layer and the hidden layer is set to the unit matrix, and its weight is not updated. The expression of the initial RBF neural network is as follows:

[0058] f(x)=W *T h(x)+ε

[0059] Where x represents the input of the RBF neural network; W * represents the ideal weight matrix between the hidden layer and the output layer in the RBF neural network, and ε is the approximation error of the RBF neural network. Under a fixed neural network, that is, under the condition of ideal weights, the approximation error of the neural network is bounded, that is, |ε|≤ε bound h(x)=[h j ] T is the Gaussian basis function output of the RBF neural network, j represents the jth node of the hidden layer of the RBF neural network, and h j The expression is as follows:

[0060]

[0061] Among them, exp(·) represents the exponential function with the natural constant e as the base, c j represents the center vector of the jth node in the network, b j Represents the basis width parameter of the j-th node and is a number greater than zero.

[0062] Since the ideal weights of the RBF neural network are unknown, the output of the RBF neural network is It is expressed as:

[0063]

[0064]

[0065] in, represents the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network, for The derivative of is an estimate of the ideal weight matrix The update law, that is, the adaptive law of the system model, is the derivative of h(x);

[0066] Therefore, the RBF neural network is used to approximate the unknown continuous-time state function of the system model, and the system model is reconstructed to obtain the reconstructed system model. Its expression is as follows:

[0067]

[0068] in, is the output of the RBF neural network, representing the unknown continuous-time state function approximated by the RBF neural network; Represents the input of the RBF neural network, which is the n-dimensional system state vector; represents the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network, T represents the transpose of the matrix, h(x) represents the output of the Gaussian basis function of the RBF neural network, is the n-dimensional system control vector.

[0069] The present invention combines the input and output data of the existing system with the RBF neural network, continuously corrects the weights of the RBF neural network to fit the functional relationship between the input variables and the output variables, minimizes the error between the predicted output and the actual output, and realizes adaptive approximation of the unknown part of the system model.

[0070] definition is the weight estimation error of the RBF neural network, then:

[0071]

[0072] In this embodiment, the last step of reconstructing the system model using the RBF neural network is to design an estimate of a suitable ideal weight matrix. The update law Its expression is as follows:

[0073]

[0074] Among them, γ is the settable learning rate, c is an adjustable parameter that needs to satisfy the Hurwitz condition, and s is the sliding mode function of the reconstructed system model.

[0075] S2: According to the reconstructed system model, a sliding mode function is constructed. Sliding mode control is essentially a control strategy of a variable structure control system. This control characteristic forces the system to move up and down with a small amplitude and high frequency along a specified state trajectory (i.e., sliding mode surface) under certain characteristics, which is the so-called sliding mode or "sliding mode" motion. This sliding mode can be designed and is independent of the parameters and disturbances of the system. Therefore, the present invention uses a sliding mode function to design a parallel controller.

[0076] The purpose of designing a parallel controller is to make the system model stable, that is, x→0, Therefore, the constructed sliding mode function s contains x and In normal operation, the general definition of the sliding mode function is:

[0077]

[0078] However, since the system model is partially unknown, that is, f(x) is unknown, the reconstructed system model is used to design the final sliding mode function s, whose expression is as follows:

[0079]

[0080] The derivative of the sliding mode function is:

[0081]

[0082] That is, the original system model is expressed as

[0083]

[0084] S3: A parallel controller is constructed based on the RBF neural network and the sliding mode function, and the parallel controller is used to control the system to run to an equilibrium state, that is, the derivative of the system state x is equal to 0. After the system runs to this state, it no longer changes and reaches stability.

[0085] In this embodiment, the parallel controller includes a switching component and a continuous component; wherein the switching component control system moves quickly toward the sliding surface, and the continuous component control system moves along the sliding surface until the system runs to a balanced state, such as Figure 3 As shown, Figure 3 is the schematic diagram of the parallel controller, and its expression is as follows:

[0086]

[0087]

[0088]

[0089] Where k>0, and

[0090]

[0091]

[0092] in, is the parallel controller, u is the control signal, represents the continuous component of the parallel controller, The switching component of the parallel controller, for Derivative of , ε bound represents the upper bound of the approximation error of the RBF neural network, sgn(·) represents the sign function, and k represents an adjustable parameter greater than zero.

[0093] The parallel controller is based on the unknown part of the system approximated by the RFF neural network and the adaptive law Generate a control signal and input the control signal into the actual system to achieve a stable control system.

[0094] Example 3

[0095] This embodiment proposes a parallel control system for partially unknown system models, such as Figure 4 As shown, Figure 4 This is an architecture diagram of a parallel control system with partially unknown system models, including: RBF neural network building module, system reconstruction module, sliding mode function design module, parallel controller building module and control module.

[0096] In the specific implementation process, the RBF neural network construction module determines the number of nodes in the input layer, hidden layer and output layer for the unknown part of the system model, selects the Gaussian basis function as the activation function, and sets the weight matrix between the input layer and the hidden layer as the unit matrix, thereby constructing the RBF neural network.

[0097] The system reconstruction module uses RBF neural network to approximate the unknown part of the system model and reconstructs the system model. By combining the existing system input and output data with the RBF neural network, the weights of the RBF neural network are continuously corrected to fit the functional relationship between the input variables and the output variables, so that the error between the predicted output and the actual output is minimized, and the adaptive approximation of the unknown part of the system model is achieved to obtain the reconstructed system model.

[0098] The sliding mode function design module constructs the sliding mode function of the reconstruction system model.

[0099] The parallel controller building module builds a parallel controller based on RBF neural network and sliding mode function.

[0100] The control module uses the parallel controller to control the system to run to a balanced state. The parallel controller includes a switching component and a continuous component; wherein the switching component control system moves quickly toward the sliding surface, and the continuous component control system moves along the sliding surface until the system runs to a balanced state, so that the closed-loop system tends to be stable.

[0101] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0102] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A parallel control method for a partially unknown system model, comprising the following steps: S1: Construct an RBF neural network, use the RBF neural network to approximate the unknown continuous-time state function of the system model, reconstruct the system model, and obtain a reconstructed system model , whose expression is as follows: in, is the output of the RBF neural network, representing the unknown continuous-time state function approximated by the RBF neural network; Represents the input of the RBF neural network, which is n dimensional system state vector; represents the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network, T represents the transpose of a matrix, represents the output of the Gaussian basis function of the RBF neural network, for n dimensional system control vector; S2: According to the reconstructed system model, a sliding mode function is constructed, and its expression is as follows: in, c is an adjustable parameter that satisfies the Hurwitz condition; S3: A parallel controller is constructed based on the RBF neural network and the sliding mode function, and the system is controlled to run to a balanced state by using the parallel controller; the parallel controller includes a switching component and a continuous component; wherein the switching component control system moves quickly to the sliding surface, and the continuous component control system moves along the sliding surface until the system runs to a balanced state, and its expression is as follows: in, ,and in, is a parallel controller, represents the continuous component of the parallel controller, The switching component of the parallel controller, for The derivative of represents the upper bound of the RBF neural network approximation error, represents the symbolic function, k is an adjustable parameter greater than zero.

2. The parallel control method for partially unknown system models according to claim 1, characterized in that: In S1, an RBF neural network is constructed based on the unknown part of the system model, including: For the unknown part of the system model, the number of nodes in the input layer, hidden layer and output layer of the RBF neural network is determined, the Gaussian basis function is selected as the activation function, and the weight matrix between the input layer and the hidden layer is set to the unit matrix to obtain the RBF neural network.

3. The parallel control method for partially unknown system models according to claim 1, characterized in that: After obtaining the sliding mode function of the reconstructed system model, it also includes designing the update law of the estimated value of the ideal weight matrix between the hidden layer and the output layer in the RBF neural network. , using the update law Update the estimate of the ideal weight matrix , the update law The expression is as follows: in, is a configurable learning rate.

4. A parallel control system for a partially unknown system model, applied to the parallel control method for a partially unknown system model as claimed in any one of claims 1 to 3, characterized in that: include: RBF neural network building module, used to build RBF neural network; A system reconstruction module, used to use the RBF neural network to approximate the unknown part of the system model, reconstruct the system model, and obtain a reconstructed system model; A sliding mode function design module, used for constructing a sliding mode function of the reconstruction system model; Parallel controller building module, used to build parallel controllers based on RBF neural network and sliding mode function; A control module is used to control the system to run to a balanced state using the parallel controller.

5. The parallel control system for partially unknown system models according to claim 4, characterized in that: The unknown part of the system model is the continuous-time state function.

6. A computer system, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program in the memory, the steps of the parallel control method for partially unknown system models as described in any one of claims 1 to 3 are implemented.

Images